Semi-automated end-to-end quality control system for annotation data based on automated Cobb angle measurement.

PURPOSE: The field of automated scoliosis diagnosis necessitates large-scale data annotation; however, existing annotation practices are hampered by the lack of a systematic, end-to-end quality control framework. To address this gap, we aim to develop a Semi-Automated End-to-End QC (SAE-QC) System designed to ensure annotation accuracy while substantially accelerating the data collection pipeline and reducing operational costs. This system is expected to provide critical support for the construction of high-quality datasets and the development of reliable AI models in the domain of automated Cobb angle measurement. METHODS: We propose the SAE-QC System for medical image annotation, featuring core modules including a data assignment module with dual-blind annotation pairing medical and non-medical annotators, an annotation protocol QC module enforcing standardized guidelines through structured validation, and an annotation QC module performing comprehensive quality checks from format validation to multi-annotator consistency analysis. The system’s modular architecture ensures cross-platform compatibility and seamless integration across the entire annotation workflow from data distribution to expert adjudication, while maintaining flexibility for different medical imaging tasks and data formats. RESULTS: We developed the SAE-QC System, comprising a Data Assignment Module, an Annotation Protocol QC Module, and an Annotation QC Module. By integrating automated algorithms with expert supervision, the system enables end-to-end management of both Cobb angle measurement and vertebral segmentation, significantly reducing error rates and improving data reliability. When applied to a dataset construction task, experimental results showed the following improvements in annotation quality: the abnormality rate for Cobb angle measurement decreased from 66% to 19%, measurement consistency improved by approximately 40%, and the abnormality rate for vertebral segmentation decreased from 27% to 2.9%, with a significant increase in overall accuracy. CONCLUSION: This study demonstrates that the SAE-QC system—integrating automated quality assurance with expert supervision—effectively enhances the reliability of medical image annotation, providing a practical framework for non-specialist teams to achieve high-quality annotations.